There are several issues involved.
- Expansion audio: You want the ratio of M2 cycles to synthesizer cycles to be the same. A synth clocked by M2 is fine. A synth clocked by its own oscillator may have problems.
- Tonal DMC samples (aka "Sunsoft bass"): You want the ratio of CPU cycles to DMC fetches to be the same.
- Raster effects: You want the ratio of PPU dots to CPU cycles to be the same, and in some cases (VRC IRQ and the cycle-timed NMI handlers of Balloon Fight and Battletoads), you want the time from NMI to start of picture to be the same.
- DMC IRQ-assisted raster effects: You want the ratio of PPU dots to DMC fetches to be the same.
- Total frame length: Most music engines use NMI as a time base for counting tempo.
An AY-3-8910 clone clocked by M2 was used in Sunsoft games that use Sunsoft bass. This means A is OK, and the limiting factor to keeping a music engine in tune on PAL NES is condition B.
The Dendy, a PAL famiclone, uses a divide-by-15 counter on the CPU, and it put the extra 50 lines of a 50 Hz TV system before NMI. This ensures condition C, and because the pitches stay almost the same, the cloner didn't have to modify the DPCM table, ensuring B, which is why Sunsoft bass should work on Dendy. Ensuring B and C ensures D, which mostly concerns compatibility with games like Time Lord and Fire Hawk.
A PAL NES, on the other hand, fails condition C because of the 2A07's divide-by-16, and to compensate, it also fails B. This means the pitch tables have to be modified to keep the rest of the music from being way flat compared to the bass. I have no access to a PAL NES and thus no way to see to what extent the changes to B and C canceled out with respect to D.
E just makes the music slower or faster. It doesn't knock things out of tune, though it might knock repeating rhythmic patterns on DMC out of sync.